io/fs: wrong comment about func Glob(fsys FS, pattern string)

[DrGo]

currently it says

// Glob ignores file system errors such as I/O errors reading directories.
// The only possible returned error is path.ErrBadPattern, reporting that
// the pattern is malformed.

but Glob now calls Stat() on pattern if pattern does not contains any of the magic characters recognized by path.Match (ie if pattern is just a file path). So any file io errors can occur (and returned).

I suspect that this change was done to allow Glob to be used in places like html/template.ParseFS which takes as arguments fs.FS and any number of glob patterns; in a way overloading it to replace both ParseGlob and ParseFiles.

I wish that ParseFS behaved more like ParseFiles (since globing is trivial to add by users if needed) or that we have both ParseGlobFS and ParseFilesFS (which IMO is cleaner: the FS versions act on fs.FS whereas the other two act on the OS filesystem). In a package I am working on, I had to copy/paste the code for template.ParseFiles into a func that takes an fs.FS because since I was processing the template files before passing them to ParseFiles, I did not need an additional Stat() call on them which would happen if I passed them to ParseFS.

Thanks,

[ianlancetaylor]

In the current code, if the call to Stat fails, Glob returns a nil error, which seems consistent with the documentation.

I am likely missing something. Can you show an example of how a problem might occur? Thanks.

[DrGo]

Sorry Ian, looked at that code for an hour but did not realize it returns nil rather than error.